Detecting Promoter Regions

Promoter regions in DNA sequences do not follow a strict pattern. This makes the identification of promoter regions more difficult. Although promoter regions vary, it is usually possible to find a DNA sequence (called the consensus sequence) to which all the of them are very similar. For example, the consensus in the bacterium E.coli, based on the study of 263 promoters, is TTGACA followed by 17 uncorrelated base pairs, followed by TATAAT, with the latter, called TATA box, located about 10 bases upstream of the transcription start site. None of the 263 promoter regions exactly match the above consensus sequence. Nevertheless, the consensus sequence is representative: nearly all of E.coli's promoters terminate with 2 of the 3 specified letters of the sequence TAxyzT, 80-90% have all 3, and xyz is TAA in approximately 50% of the promoter regions. Due to the high variability, exact methods cannot be used for identifying promoter regions by the TATA box. Instead, we use a pattern search method based on frequencies. We construct a table of statistics, f_b,i, where f_b,i is the frequency of the base b in position i of the known promoter region suffixes. We assume positions are independent. Let f_b denote the expected frequency of the base b in the genome. We calculate the likelihood of a given sequence being a TATA-box. For a sequence $S=B_1B_2\ldots B_6$ the likelihood of it being a TATA-box is:

$$P(S\vert S\mbox{ is a TATA-box})=\prod_{i=1}^{6}f_{B_i,i}$$

Similarly, the likelihood of observing it, given it is a "non-promoter" is:

$$P(S\vert S\mbox{ is not a TATA-box})=\prod_{i=1}^{6}f_{B_i}$$

The log-likelihood ratio is therefore:

$$\log\left(\frac{P(S\vert\mbox{promoter})}{P(S\vert\mbox{non-p... ...ht)= \sum_{i=1}^{6}\log\left(\frac{f_{B_i,i}}{f_{B_i}}\right)$$

From the table f_Bi,i we therefore construct a scoring matrix, with each entry S_b,i denoting the score that a sequence should be given for having the base b in the i-th position. The score s_b,i is computed by the following formula:

$$s_{b,i}=\log \left( \frac {f_{b,i}}{f_b} \right)$$

The algorithm simply scans through the DNA sequence and computes the likelihood ratio for every consecutive 6 bases and thus locates regions where the likelihood ratio is high. This model has the disadvantage that it doesn't exploit all of the known information (i.e. intron/exon statistic, dependencies between bases occurring in the promoter regions etc.) Why aren't promoters precise like the stop codons etc.? A likely answer is that nature uses the variation in promoters to control expression levels of various genes. That is, the rate of the gene expression process depends on the conservation of the promoter region. This hypothesis is supported by results from chemistry. Experiments show that when a RNA polymerase molecule gets bounded to the prompter region in order to initial transcription, there is an 80% correlation between the weight matrix score of the region and the binding energy. This means that if the promoter region is very conserved, i.e., very similar to the consensus sequence, then the binding energy barrier is low and thus the protein production rate is higher (because the RNA polymerase can easily bind to the protein coding region). When the difference from the consensus sequence is bigger, the energy barrier is higher, and the protein production is slower. The unfortunate consequence is that rarely expressed genes will be harder to find by this means.

  

Figure 7.8: Promoter detection by matrix evaluations of the sequence for containing a TATA-box. The matrix contains an element for each possible matrix position. For each alignment of the matrix above the sequence, a score is computed based on the matrix elements corresponding to the sequence. The matrix rows correspond to the bases A,C,G,T from top to bottom. The sequence TATAAT scores highest.